Saturday 01 February 2025
The quest for a reliable way to predict long-term complications in patients with type 2 diabetes has taken another step forward, thanks to a team of researchers who have developed a new approach using natural language processing techniques.
Traditionally, doctors rely on clinical codes and patient records to identify individuals at risk of developing microvascular complications such as retinopathy, nephropathy, and neuropathy. However, these methods often struggle to accurately predict the likelihood of these complications occurring in individual patients.
The researchers, led by Elizabeth Remfry and Rafael Henkin from Queen Mary University of London, have developed a code-agnostic approach that uses textual descriptors associated with clinical codes to identify patients at risk of developing microvascular complications.
Their method involves encoding individual electronic health records (EHRs) as text using a fine-tuned language model, and then training a neural network on this data to predict the likelihood of microvascular complications occurring in individual patients. The researchers tested their approach on a large dataset of EHRs from over 133,000 patients with type 2 diabetes.
The results show that the code-agnostic approach outperforms traditional code-based methods in predicting long-term microvascular complications, particularly when considering longer timeframes such as five and ten years. The model also performed better at predicting the risk of retinopathy, the most common first complication to occur in patients with type 2 diabetes.
One of the key advantages of the new approach is its ability to capture complex relationships between different clinical codes and patient characteristics. By using textual descriptors, the researchers can tap into existing clinical knowledge and incorporate information from multiple sources, including diagnoses, symptoms, prescriptions, and referrals.
The study’s findings have significant implications for healthcare practitioners who are working to develop more effective strategies for predicting and preventing microvascular complications in patients with type 2 diabetes. The code-agnostic approach could potentially be applied to other chronic conditions, such as heart disease or cancer, where accurate prediction of long-term outcomes is critical.
In addition to its potential clinical applications, the study highlights the importance of developing more sophisticated machine learning models that can handle complex healthcare data. As electronic health records continue to grow in size and complexity, researchers will need to develop new approaches that can effectively harness this data to improve patient care.
Cite this article: “Predicting Long-Term Complications in Type 2 Diabetes Using Natural Language Processing”, The Science Archive, 2025.
Type 2 Diabetes, Natural Language Processing, Microvascular Complications, Electronic Health Records, Neural Network, Code-Agnostic Approach, Machine Learning, Patient Records, Chronic Conditions, Predictive Modeling
